During a two-day trip I ended up building a 6-agent conversational OS with GPT-5.1 — using only natural-language instructions, no code.
It’s called PolyAgora.
What it is:
– A natural-language multi-agent “OS” with 6 agents
– Built around a tri-axis core (Arc–Ann–Saku) + 3 supporting agents
– Designed to produce long-form, philosophy-grade conversations
– Dynamic objection engine + topic-shift mechanics
– Emergent behavior, not pre-scripted roleplay
I’m not an ML engineer.
This grew organically out of wanting to have deep small-talk with GPT-5.1 during a short trip — and it evolved into a structured OS.
Repo (docs + long examples):
https://github.com/Takeshi-Sakamoto5/PolyAgora
I’d love feedback from HN:
– Is this structure useful for multi-agent reasoning research?
– Does the emergent conversation quality seem meaningful?
– What would you add/change for v1.1.0 (I plan to include a “runaway dialogue” example)?
Comments URL: https://news.ycombinator.com/item?id=45946621
Points: 1
# Comments: 0
Source: github.com
